Floating point values are stored using the IEEE754 Single Precision 4-byte format in little endian byte order.

Where:

Exp: 8-bit exponent stored with an offset bias of 127. The values 00 and FF have unique meanings.

Fract: 23-bit fraction. If the exponent is > 0, then the mantissa is 1.fract. If the exponent is zero, then the mantissa is 0.fract.

The floating point value depends on the unique cases of the exponent:

• If the exponent is FF and the fraction is zero, this represents ± infinity.
• If the exponent is FF and the fraction is non-zero this represents "not a number" (NaN).
• If the exponent is 00 then the value is a subnormal number represented by (–1)^sign × 2^–126 × 0.fraction.
• Otherwise, the value is a normalized number represented by (–1)^sign × 2^{(exponent – 127)} × 1.fraction.